Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Cannot Reproduce
-
N/A
-
None
Description
perl /home/elenst/git/rqg/runall-new.pl --no-mask --seed=1476316781 --threads=32 --duration=600 --queries=100M --mysqld=--log_output=FILE --mysqld=--innodb-buffer-pool-size=2G --mysqld=--innodb-log-file-size=256M --mysqld=--innodb-flush-log-at-trx-commit=2 --grammar=conf/partitioning/partition_pruning.yy --gendata=conf/partitioning/partition_pruning.zz --mtr-build-thread=72 --basedir=/home/elenst/git/10.2-rel --vardir=/home/elenst/test_results/analyze
|
The test sporadically (but frequently enough) fails like this:
# 2016-10-14T10:30:24 [5916] Mixer: Server crash or critical failure (STATUS_SERVER_CRASHED) reported at dsn dbi:mysql:host=127.0.0.1:port=10720:user=rqg:database=test:
|
mysql_local_infile=1
|
# 2016-10-14T10:31:07 [5942] GenTest: Child worker process completed successfully.
|
# 2016-10-14T10:31:07 [5942] GenTest: child is being stopped with status STATUS_OK
|
# 2016-10-14T10:31:45 [5916] GenTest: Server crash or critical failure (STATUS_SERVER_CRASHED) reported, the child will be stopped
|
# 2016-10-14T10:31:45 [5916] GenTest: child is being stopped with status STATUS_SERVER_CRASHED
|
# 2016-10-14T10:33:55 [5813] Test completed with failure status STATUS_SERVER_CRASHED (101)
|
# 2016-10-14T10:33:55 [5813] GenTest exited with exit status STATUS_SERVER_CRASHED (101)
|
There is, however, no sign of crash in the server error log.
Here is the last query that the connection executed:
16 Query SELECT table1.`col_varchar_256_utf8_key` AS field1 FROM `PP_J` AS table1 RIGHT OUTER JOIN `PP_P` AS table2 ON ( table2.`col_int_signed_k
|
ey` = table1.`col_int_signed_key` ) WHERE NOT ( table1.`col_date` = '2000-05-15' ) /* QNO 2523 CON_ID 16 PID 5916 */
|
It can be different, but so far it has always been a select with a LEFT/RIGHT join. Maybe just because it's slow.
There is no KILL in the test flow. Adding skip-shutdown does not help, so it's not a connection dying on server shutdown, either.
It might well be a bogus error on the test side, but it needs to be investigated.